Patient-provider communication impacts healthcare outcomes, but assessing the quality of interactions manually takes time and effort. This project explores the automatic assessment of patient-provider interactions using Language Style Matching (LSM). LSM scores the linguistic similarity of function words between conversational partners (e.g.,  pronouns, articles) from 0 (low matching) to 1 (perfect matching), reflecting how in-sync partners are. Past research establishes LSM as a marker for the quality of interpersonal communication that predicts how likely romantic relationships are to last, but has not been explored for clinical interactions. We (CL, NE) applied LSM to investigate how well patients and providers matched each other's speaking styles for insights into the quality of clinical interactions. We used Linguistic Inquiry and Word Count (LIWC), a software program for LSM analysis of transcripts. Using LIWC, we analyzed the transcripts of 108 simulated visits between 54 primary care providers and four standardized patients. We used descriptive statistics to characterize LSM across visits. Our initial findings show that LSM scores range from 0.77 to 0.94 ( mean=0.86, SD=0.03) which is similar to prior research where most verbal conversations fall between 0.83 and 0.94. These findings show that on average providers and patients tend to match each other in their speaking style at a level similar to typical conversations. However, we identified some outliers that fall below 0.83 threshold. Opportunities for future work include thin-slice analysis of the transcripts to understand how LSM scores change throughout a visit and comparing LSM scores to self-reported survey data about visit quality. ​​We hope to further investigate this efficient marker of conversational quality as LSM has the potential to characterize the quality of clinical interactions without the time and effort required of traditional manual approaches.

The Svoboda Diaries Project (SDP) is an interdisciplinary digital humanities effort dedicated to preserving and analyzing a collection of historical diaries. Written by Joseph Svoboda, a British steamship purser, these diaries span 40 years and offer firsthand insights into the daily life, healthcare practices, social networks, economic conditions, and cultural landscape of 19th century Iraq. Due to the unstructured and handwritten nature of these texts, analysis of them could benefit from computational techniques. Our research specifically addresses this challenge by extracting and analyzing references to food, medicine, symptoms, and healthcare providers, making these diaries a valuable resource for studying historical medical practices. To extract key terms from the diaries, we apply Natural Language Processing (NLP) methods,  which allow computers to interpret human language. Specifically, we use Named Entity Recognition (NER), a technique that identifies and categorizes entities or terms such as foods, medicines, illnesses, and doctors within the diaries. This type of extraction allows us to transform narrative into a format that is more amenable to analysis using automated methods. Once extracted, we visualize these relationships through network visualizations—graphical representations that illustrate connections between different entities in the text. These visualizations help us trace the circulation of medical knowledge, showing who prescribed what, which remedies were most common in the diaries, and how treatment preferences may have varied depending on provider perspectives. We aim to directly link healthcare providers with the remedies they recommended, allowing us to understand patterns of medical practice at the time. Structuring historical information into data-driven models allows us to examine cultural and economic influences on healthcare. Beyond this specific case, our research demonstrates how data science can be applied to historical texts, enabling researchers to discover patterns in historical healthcare practices across different time periods and regions.

Stigma related to substance use disorders (SUDs) has profound and far-reaching consequences on individuals’ physical and mental health, as well as their socioeconomic well-being. This can lead to social isolation and can hinder access to treatment. While most stigma reduction interventions target structural stigma (such as educating medical students or professionals who work with individuals with SUDs), our study targets social and self-stigma experienced by individuals ages 21-35 with risky alcohol and/or cannabis use. Our objective is to develop a digital intervention to support individuals who experience substance use related stigma, enabling them to cope with stigma more effectively. Its digital nature allows for increased accessibility, convenience, and consistent support to individuals who might otherwise face barriers in accessing traditional healthcare services. We employ a user-centered design approach, utilizing peer mentoring to reduce self-stigma. Over 3 Zoom sessions, we collaborate with participants to first brainstorm topics that should be included in the intervention, receive feedback on different activities participants might engage in, and finally gather feedback on a prototype of the intervention which includes completing the System Usability Scale. My role involves collaboratively drafting the focus group scripts and facilitating sessions, including engaging participants in activities. From the focus groups, we anticipate that we will better understand important factors that influence the effectiveness of the intervention such as what features are most engaging, what format of content is most effective, and what topics are most relevant to the population. Findings from this project will allow us to design an effective digital intervention that can reach more people and provide an alternative for those who may not be willing or able to access the healthcare system.